Code:
<span class="sigil_index_marker"><i>Silent Fail</i></span>
Inclusion of html tags as above causes a silent fail, where the item is ignored with no indication to the author. In engineering, this is he worst kind of failure.
Why not have the index generator just ignore <i>, <b>, and <u> as html tags? Don't strip them, just special-case those three simple tags and leave them be. That doesn't address the broader problem, but it does at least partially turn a bug into a feature - it would allow an author to pass some simple formatting into the index.
While editing, I use css color highlighting for sigil_index_marker. That makes it much easier to see what is indexed in the preview, and can be turned off for publication by commenting-out the css rule. That functionality might not be a bad thing to build into the program itself, with a button to turn it on and off. (Also sigil_split_marker)
My index has about 1000 entries so far, and editing it by hand after every time it is regenerated is completely impractical. About a quarter of it is scientific nomenclature, ie Latin this is written in italics by convention.